This vignette explains how to performs ionomics data analysis including gene
network and enrichment analysis by using the modification of R package,
ionflow. The
modification(ionflow_funcs) was made by Wanchang Lin
(w.lin@imperial.ac.uk) and Jacopo Iacovacci (j.iacovacci@imperial.ac.uk).
To explore the pipeline, we’ll use the ionomics data set:
ion_data <- read.table("../test-data/iondata.tsv", header = T, sep = "\t")
dim(ion_data)
#> [1] 9999 16
Ten random lines are shown as:
sample_n(ion_data, 10)
| Knockout | Batch_ID | Ca | Cd | Co | Cu | Fe | K | Mg | Mn | Mo | Na | Ni | P | S | Zn |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| YGR197C | 16 | 135.26 | 0.93 | 0.14 | 1.82 | 10.31 | 3298.19 | 863.75 | 1.44 | 2.78 | 275.31 | 0.96 | 5760.13 | 795.04 | 26.44 |
| YDL227C | 6 | 20.35 | 0.67 | 0.11 | 0.89 | 1.50 | 1441.82 | 264.75 | 0.68 | 0.44 | 76.36 | 0.31 | 1693.47 | 61.27 | 12.22 |
| YLR181C | 21 | 45.94 | 1.23 | 0.20 | 2.00 | 9.77 | 3133.33 | 590.65 | 1.65 | 1.00 | 216.31 | 8.83 | 4886.48 | 438.14 | 17.99 |
| YDL227C | 89 | 31.76 | 0.88 | 0.14 | 1.46 | 7.90 | 2086.22 | 662.71 | 0.94 | 1.07 | 190.91 | 1.68 | 4141.19 | 429.41 | 11.75 |
| YJL192C | 28 | 45.49 | 1.55 | 0.17 | 1.75 | 9.22 | 2606.80 | 681.01 | 1.55 | 1.30 | 217.81 | 1.62 | 4814.28 | 596.28 | 16.48 |
| YLR133W | 21 | 36.90 | 0.90 | 0.19 | 1.95 | 10.25 | 2942.86 | 600.78 | 1.30 | 1.10 | 181.62 | 1.40 | 4492.62 | 489.04 | 15.79 |
| YKL106W | 23 | 36.73 | 0.99 | 0.20 | 1.71 | 7.60 | 2301.65 | 647.38 | 0.97 | 0.77 | 157.70 | 1.45 | 4320.44 | 442.54 | 15.19 |
| YDL227C | 91 | 22.14 | 0.76 | 0.11 | 0.85 | 5.60 | 2706.70 | 672.84 | 0.88 | 1.30 | 203.78 | 0.56 | 4277.77 | 467.11 | 13.96 |
| YJL144W | 29 | 47.58 | 1.17 | 0.16 | 1.82 | 9.66 | 2854.36 | 769.41 | 1.25 | 1.23 | 250.44 | 1.64 | 5149.42 | 555.72 | 16.90 |
| YLR165C | 80 | 13.23 | 1.02 | 0.15 | 3.10 | 7.73 | 2563.36 | 804.38 | 0.96 | 1.55 | 237.82 | 1.30 | 5634.82 | 448.21 | 16.36 |
The first few columns are meta information such as gene ORF and batch id. The rest is the ionomics data.
The raw data set should be pre-processed. The pre-processing function
PreProcessing performs:
The raw data are at first log trainsformed and then followed by the batch
correction. The user can chose not to perform batch correction, otherwise
the user can use either median or median plus std method. If there is
quality control for the batch correction, the user can use it and indicates
in the argument of control_lines. Also this function gives user option how
to use these control line (control_use): If control_use is control,
these control lines (data rows) are used for the batch correction factor; if
control.out, lines except control lines are used.
This data set has a control line: YDL227C mutant. The code segment below is to identify it:
max(with(ion_data, table(Knockout)))
#> [1] 1617
which.max(with(ion_data, table(Knockout)))
#> YDL227C
#> 209
The next stage is outlier detection. Here only univarite methods are
implemented, including mad, IQR, and log.FC.dist. And like batch
correction, user can skip this procedure by setting method_outliers = none
in the function argument. There is a threshold to control the number of
outliers. The larger the threshold (thres_outl) the more outlier removal.
Standarisation provides three methods: std, mad or custom. If the method is cumstom, user must use specific std values such as:
std <- read.table("../test-data/user_std.tsv", header = T, sep = "\t")
std
#> Ion sd
#> 1 Ca 0.1508
#> 2 Cd 0.0573
#> 3 Co 0.0580
#> 4 Cu 0.0735
#> 5 Fe 0.1639
#> 6 K 0.0940
#> 7 Mg 0.0597
#> 8 Mn 0.0771
#> 9 Mo 0.1142
#> 10 Na 0.1075
#> 11 Ni 0.0784
#> 12 P 0.0597
#> 13 S 0.0801
#> 14 Zn 0.0671
The pre-process procedure returns not only processed ionomics data but also
a symbolic data set. This data set is based on the inomics data and is
determined by a threshold(thres_symb):
0 if ionomics value is located between [-thres_symb, thres_symb]1 if ionomics value is larger than thres_symb-1 if ionomics value is smaller than -thres_symbThe core part of network and enrivhment analysis, clustering, is based on the symbolic data.
Let’s run the pre-process procedure:
pre <- PreProcessing(data = ion_data,
var_id = 1, batch_id = 2, data_id = 3,
method_norm = "median",
control_lines = "YDL227C",
control_use = "control",
method_outliers = "IQR",
thres_outl = 3,
stand_method = "std",
stdev = NULL,
thres_symb = 3)
names(pre)
#> [1] "stats.raw_data" "stats.outliers" "stats.batch_data"
#> [4] "data.long" "data.gene.logFC" "data.gene.zscores"
#> [7] "data.gene.symb" "plot.dot" "plot.hist"
The results includes summaries of raw data and processed data. The latter is:
pre$stats.batch_data %>%
kable(caption = 'Processed data summary', digits = 2, booktabs = T) %>%
kable_styling(full_width = F, font_size = 10)
| Ion | Min. | 1st Qu. | Median | Mean | 3rd Qu. | Max. | Variance |
|---|---|---|---|---|---|---|---|
| Ca | -4.45 | -0.28 | -0.13 | -0.12 | 0.02 | 2.35 | 0.11 |
| Cd | -1.70 | 0.03 | 0.10 | 0.11 | 0.17 | 0.93 | 0.03 |
| Co | -2.80 | 0.02 | 0.09 | 0.06 | 0.15 | 1.60 | 0.05 |
| Cu | -0.66 | -0.10 | -0.03 | -0.01 | 0.04 | 5.28 | 0.04 |
| Fe | -7.48 | -0.17 | -0.06 | -0.02 | 0.07 | 6.88 | 0.14 |
| K | -2.21 | -0.17 | -0.01 | -0.08 | 0.09 | 1.83 | 0.08 |
| Mg | -1.84 | -0.06 | 0.01 | -0.01 | 0.07 | 1.69 | 0.03 |
| Mn | -4.11 | -0.24 | -0.08 | -0.13 | 0.01 | 1.78 | 0.06 |
| Mo | -2.03 | -0.26 | -0.08 | -0.08 | 0.09 | 4.44 | 0.13 |
| Na | -7.41 | -0.53 | -0.22 | -0.33 | -0.04 | 1.25 | 0.24 |
| Ni | -2.40 | -0.01 | 0.09 | 0.12 | 0.21 | 7.90 | 0.12 |
| P | -1.18 | -0.06 | 0.00 | -0.01 | 0.06 | 1.45 | 0.02 |
| S | -2.38 | -0.03 | 0.05 | 0.06 | 0.16 | 2.38 | 0.04 |
| Zn | -0.46 | -0.08 | -0.03 | -0.01 | 0.03 | 4.60 | 0.02 |
The pre-processed data and symbolic data are like like:
pre$data.gene.zscores %>% head() %>%
kable(caption = 'Processed data', digits = 2, booktabs = T) %>%
kable_styling(full_width = F, font_size = 10,
latex_options = c("striped", "scale_down"))
| Line | Ca | Cd | Co | Cu | Fe | K | Mg | Mn | Mo | Na | Ni | P | S | Zn |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| YAL004W | -1.16 | 0.75 | 1.19 | -0.47 | 0.04 | 0.61 | 0.51 | -0.84 | -0.08 | -1.84 | 1.71 | 0.52 | 0.33 | -0.09 |
| YAL005C | -1.67 | 0.84 | 0.55 | 0.58 | -2.79 | 0.59 | 0.31 | -1.16 | -1.42 | -0.12 | 1.48 | 0.73 | 0.13 | -0.13 |
| YAL007C | -2.12 | 0.64 | 0.23 | -0.53 | -0.24 | 0.79 | -0.09 | -0.14 | 1.22 | -0.92 | 0.00 | 0.09 | -0.29 | -0.65 |
| YAL008W | -2.34 | 1.13 | 0.21 | -0.73 | -2.16 | 0.52 | -0.02 | -0.87 | 0.93 | -0.58 | 0.02 | -0.09 | -0.73 | -0.47 |
| YAL009W | -1.18 | 0.66 | 0.55 | -1.11 | -3.91 | 0.22 | 0.09 | -0.18 | 1.50 | -0.84 | -0.09 | 0.14 | 0.01 | -0.36 |
| YAL010C | -1.28 | 1.43 | 2.27 | 0.46 | 1.53 | -2.75 | 0.04 | -0.74 | -9.71 | -4.30 | 2.42 | -0.98 | -0.05 | -0.01 |
pre$data.gene.symb %>% head() %>%
kable(caption = 'Symbolic data', booktabs = T) %>%
kable_styling(full_width = F, font_size = 10)
| Line | Ca | Cd | Co | Cu | Fe | K | Mg | Mn | Mo | Na | Ni | P | S | Zn |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| YAL004W | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL005C | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL007C | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL008W | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL009W | 0 | 0 | 0 | 0 | -1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL010C | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | -1 | -1 | 0 | 0 | 0 | 0 |
The symbolic data are calulated from the processed data with control of
thres_symb (here is 3). You can obtain a new symbol data set by
re-assigning a new threshold to the function symbol_data:
data_symb <- symbol_data(pre$data.gene.zscores, thres_symb = 2)
data_symb %>% head() %>%
kable(caption = 'Symbolic data with threshold of 2', booktabs = T) %>%
kable_styling(full_width = F, font_size = 10)
| Line | Ca | Cd | Co | Cu | Fe | K | Mg | Mn | Mo | Na | Ni | P | S | Zn |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| YAL004W | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL005C | 0 | 0 | 0 | 0 | -1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL007C | -1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL008W | -1 | 0 | 0 | 0 | -1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL009W | 0 | 0 | 0 | 0 | -1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| YAL010C | 0 | 0 | 1 | 0 | 0 | -1 | 0 | 0 | -1 | -1 | 1 | 0 | 0 | 0 |
The pre-processed data distribution is:
pre$plot.hist
Figure 1: Ionomcs data distribution plot
There are a lot of ways to filter genes. Here we filter genes based on symbolic data: remove genes with all velues are zero.
data <- pre$data.gene.zscores
data_symb <- pre$data.gene.symb
idx <- rowSums(abs(data_symb[, -1])) > 0
dat <- data[idx, ]
dat_symb <- data_symb[idx, ]
dim(dat)
#> [1] 549 15
The hierarchical cluster analysis is the key part of gene network and gene enrichment analysis. The methodology is as follow:
One example is:
clust <- gene_clus(dat_symb[, -1], min_clust_size = 10)
names(clust)
#> [1] "clus" "idx" "tab" "tab_sub"
The cluster centres are:
clust$tab_sub
#> cluster nGenes
#> 1 4 149
#> 2 11 72
#> 3 7 36
#> 4 1 27
#> 5 18 15
#> 6 5 12
#> 7 3 11
#> 8 8 11
It indicates that clusters and their number of genes (larger than
min_cluster_size).
The gene network uses both the ionomics and symboloc data. The similarity measures on the ionomics data are filtered by the similarity threshold located between 0 and 1, and cluster centres of symbolic data. The filter values are then used for network analysis.
The similarity measure method is one of pearson, spearman, kendall, cosine, mahal_cosine or hybrid_mahal_cosine. For the last two methods, see publication: Extraction and Integration of Genetic Networks from Short-Profile Omic Data Sets for details.
For example, we use the Pearson correlation as similarity measure for netwok analysis:
net <- GeneNetwork(data = dat,
data_symb = dat_symb,
min_clust_size = 10,
thres_corr = 0.75,
method_corr = "pearson")
The network with nodes coloured by the symbolic data clustering is:
net$plot.pnet1
Figure 2: Netwok analysis based on Pearson correlation: symbolic clustering
The same network, but nodes are colured by the netwok community detection:
net$plot.pnet2
Figure 3: Netwok analysis based on Pearson correlation: community detction
The network analysis also returns a network impact and betweeness plot:
net$plot.impact_betweenness
Figure 4: Netwok analysis based on Pearson correlation: impact and betweeness
For the comparision purpose, we use different similarity methods. Here we choose Cosine:
net_1 <- GeneNetwork(data = dat,
data_symb = dat_symb,
min_clust_size = 10,
thres_corr = 0.75,
method_corr = "cosine")
net_1$plot.pnet1
Figure 5: Netwok analysis based on Cosine
net_1$plot.pnet2
Figure 6: Netwok analysis based on Cosine
Use Hybrid Mahalanobis Cosine:
net_2 <- GeneNetwork(data = dat,
data_symb = dat_symb,
min_clust_size = 10,
thres_corr = 0.75,
method_corr = "mahal_cosine")
net_2$plot.pnet1
Figure 7: Netwok analysis based on Mahalanobis Cosine
net_2$plot.pnet2
Figure 8: Netwok analysis based on Mahalanobis Cosine
Again, we use Hybrid Mahalanobis Cosine:
net_3 <- GeneNetwork(data = dat,
data_symb = dat_symb,
min_clust_size = 10,
thres_corr = 0.75,
method_corr = "hybrid_mahal_cosine")
net_3$plot.pnet1
Figure 9: Netwok analysis based on Hybrid Mahalanobis Cosine
net_3$plot.pnet2
Figure 10: Netwok analysis based on Hybrid Mahalanobis Cosine
The KEGG enrichment analysis:
kegg <- kegg_enrich(data = dat_symb, min_clust_size = 10, pval = 0.05,
annot_pkg = "org.Sc.sgd.db")
#' kegg
kegg %>%
kable(caption = 'KEGG enrichmenat analysis', digits = 3, booktabs = T) %>%
kable_styling(full_width = F, font_size = 10,
latex_options = c("striped", "scale_down"))
| Cluster | KEGGID | Pvalue | Count | Size | Term |
|---|---|---|---|---|---|
| Cluster 7 (36 genes) | 03010 | 0.029 | 9 | 16 | Ribosome |
| Cluster 7 (36 genes) | 00330 | 0.031 | 3 | 3 | Arginine and proline metabolism |
| Cluster 18 (15 genes) | 00290 | 0.009 | 2 | 2 | Valine, leucine and isoleucine biosynthesis |
| Cluster 18 (15 genes) | 00520 | 0.009 | 2 | 2 | Amino sugar and nucleotide sugar metabolism |
| Cluster 18 (15 genes) | 00260 | 0.012 | 3 | 6 | Glycine, serine and threonine metabolism |
| Cluster 18 (15 genes) | 00010 | 0.024 | 2 | 3 | Glycolysis / Gluconeogenesis |
| Cluster 18 (15 genes) | 01110 | 0.037 | 5 | 22 | Biosynthesis of secondary metabolites |
| Cluster 3 (11 genes) | 00400 | 0.009 | 2 | 2 | Phenylalanine, tyrosine and tryptophan biosynthesis |
| Cluster 8 (11 genes) | 01100 | 0.006 | 6 | 55 | Metabolic pathways |
| Cluster 8 (11 genes) | 00564 | 0.027 | 2 | 6 | Glycerophospholipid metabolism |
Note that there can be none results for KRGG enrichment analysis. Change
arguments such as thres_clus as appropriate.
The GO Terms enrichment analysis:
go <- go_enrich(data = dat_symb, min_clust_size = 10, pval = 0.05,
ont = "BP", annot_pkg = "org.Sc.sgd.db")
#' go
go %>% head() %>%
kable(caption = 'GO Terms enrichmenat analysis', digits = 3, booktabs = T) %>%
kable_styling(full_width = F, font_size = 10,
latex_options = c("striped", "scale_down"))
| Cluster | ID | Description | Pvalue | Count | CountUniverse | Ontology |
|---|---|---|---|---|---|---|
| Cluster 4 (149 genes) | GO:0051336 | regulation of hydrolase activity | 0.0018 | 4 | 12 | BP |
| Cluster 4 (149 genes) | GO:0043085 | positive regulation of catalytic activity | 0.0044 | 4 | 15 | BP |
| Cluster 4 (149 genes) | GO:0035303 | regulation of dephosphorylation | 0.0068 | 2 | 3 | BP |
| Cluster 4 (149 genes) | GO:0046889 | positive regulation of lipid biosynthetic process | 0.0068 | 2 | 3 | BP |
| Cluster 4 (149 genes) | GO:1903727 | positive regulation of phospholipid metabolic process | 0.0068 | 2 | 3 | BP |
| Cluster 4 (149 genes) | GO:0044764 | multi-organism cellular process | 0.0074 | 3 | 9 | BP |
Some analysis are performed in terms of ions, i.e. feature, including PCA and correlation.
expl <- ExploratoryAnalysis(data = dat)
Figure 11: Exploratory analysis plots with respect to ionome
Figure 12: Exploratory analysis plots with respect to ionome
Figure 13: Exploratory analysis plots with respect to ionome
expl$plot.PCA_Individual
Figure 14: Exploratory analysis plots with respect to ionome
expl$plot.correlation_network
Figure 15: Exploratory analysis plots with respect to ionome